Pattern Recognition Applet: Class Dependent Linear Discrimination Analysis

The LBG-Clustering algorithm uses the following procedure to determine the line of discrimination between the data sets:

First, set the number of iterations (default = 10) from the Edit->Settings menu. For this exercise, we will use the default value.
Initially all data that is entered is pooled together into one large cluster. After creating the initial cluster, a centroid is generated be computing the mean of the initial cluster. Once the cluster and centroid have been generated classification begins. Classification involves iterating over all points in the cluster and creating new clusters based on their proximity to the centroid/centroids.
After the new clusters are generated the old centroid/centroids is/are replaced with new ones. The new centroids are generated by first computing the means of the new clusters. The new centroids are then determined by taking points which are one standard deviation towards the left and one standard deviation towards the right of the means.
The whole process above is repeated N times (N is the number of iterations) with the intent that eventually the line of discrimination between the data sets will converge.

Click here to go back to the main tutorial page.